Weee... My did ran an analysis yesterday. Unfortunately, somehow one 
sample got missed out. So now my dad has to analyse this one sample that 
got missed.

"Big deal", you say. "All you have to do is put the test tube in the 
machine and write down the number it says, right?"

Er, no. It doesn't work like that. (Which is basically why the company I 
work for even *exists* to start with, BTW.)

In fact, my dad has to configure the machine to perform *thirty four* 
seperate injections, only one of which is the sample to actually be 
measured! :-D

This, truly, is an example of a process where the setup cost is 
everything. "Normally" you'd be analysing at least 20 samples, probably 
more like 200. In that case, the 30 extra solutions you have to mix up 
would be nothing. But when you get down to analysing one lone sample... 
well it gets expensive to pay somebody to spend an hour setting up a 
machine to produce one number, let's say.

To understand why these 30 extra injections are there, you need to 
understand how the machine works. The "machine" itself uses a robotic 
arm to "inject" the liquid into the system, and it generates a signal 
that the computer records. The computer then calculates a number from this.

So for each injection, you get a number. What you need to figure out is 
what the hell that number means. This, unfortunately, changes all the 
time, so you can't just hard-code it into the computer. All kinds of 
factors can affect it. The machine can also be subject to contamination 
from other analysis that has been performed with it, and so forth.

In short, each time you want to use the machine, you have to run a whole 
set of solutions in addition to the actual samples. These are used to 
check that the machine is behaving in a consistent way, and to generate 
a calibration curve. The numbers produce go through all sorts of 
statistical acceptance tests before being used to compute actual 
chemical concentrations.

For example, you do a full calibration curve every X samples, and a 
minimum of 2 in total. If the curves don't match each other closely 
enough, the machine's response has "drifted" during the course of the 
analysis, for whatever reason. Clearly, if the calibration curves don't 
agree, you can't reliably compute any concentrations.

...and *that* is why you need 34 injections to analyse one sample! ;-)

Post a reply to this message